Cartographer README

MCP governance server in Go for the Agentic Wiki — knowledge that composes, not that you query.

[!WARNING] Beta software. Cartographer is pre-1.0: the MCP tool surface, CLI and configuration may change between minor releases without a deprecation period. Breaking changes bump the minor version (0.x semantics) and are called out in the changelog. Expect rough edges — bug reports are very welcome.

LLM agents forget everything between sessions, and stateless RAG only bolts retrieval onto that amnesia. The alternative is a knowledge base the agent itself builds and maintains over time — but letting an agent loose on a folder of files ends in broken links, lost history, and silent corruption. Cartographer is the governance layer that makes the pattern safe: the agent works the wiki exclusively through MCP tools, and the server enforces every invariant — validation, linking, immutability gates, one git commit per write.

_{The embedded Atlas on a generated demo KB (web/scripts/demo-kb.mjs): a selected concept in light, the overview in dark.}

What is it

Cartographer implements the Agentic Wiki: a persistent knowledge base of interlinked Markdown files that an LLM agent grows and curates by talking to the server over the MCP protocol. The agent never touches the files directly.

The wiki is grounded in Karpathy's "LLM Wiki" pattern (operating model: knowledge accretes over time, it is not stateless RAG) on top of the OKF substrate (Open Knowledge Format v0.1 by Google Cloud) — each KB is a folder of .md files with YAML frontmatter, self-contained and version-controlled with git. Zero lock-in: the wiki is readable by any tool, including Obsidian and any text editor.

One binary, two transports — deployment choices, not separate products: the KB model and the MCP tools are the same on both.

• Local stdio — one client, one KB, no network. The simplest way to run the pattern.
• HTTP server — one or more KBs behind bearer-token auth, as a native user service, a container or a Kubernetes workload. It hands each client the artifacts its KBs define, and lets a team share some KBs while keeping others private.

Two consequences are worth stating on their own, because they are what most of the design is for: the KB configures the agents that read it across every client you use, and it does that for a whole team rather than a single laptop.

Quick start

You need git and an empty git repository you own (GitHub, Gitea, any git host) to be your first KB's remote: a KB is a git repository, and that remote is what makes it durable and syncable. sops in PATH is needed only if the KB will hold encrypted values.

brew install beppetemp/tap/cartographer   # macOS; Windows, Linux and source installs — see Install
cartographer setup                        # server, first KB and agents, in one guided step

setup asks for the remote, shows its plan, and only then runs it: it installs the native service, creates the KB in an empty repository or mounts the one a repository already holds, connects every detected agent client and verifies the server. It checks git and the remote's credentials before changing anything, and a re-run skips what is already done. Unattended: cartographer setup --remote <url> --agents claude --yes. Each step is also its own command (service install, kb create/kb clone, connect) — see docs/configurator.md. Agent sessions that were already open need a restart to see the new MCP server.

Once connected:

cartographer status    # drift check and client/server version check; exit 0 in-sync / 1 drift / 2 error
cartographer sync      # re-apply after drift
cartographer doctor    # read-only diagnosis of the client configuration: residues, drift, missing triggers
cartographer reconnect # rebuild a client configuration from scratch, preserving every setting

The step-by-step walkthrough is docs/getting-started.md. For a single KB over stdio with no service (typically development) or a hand-configured HTTP server, see serve in docs/deployment.md.

Installing Cartographer with an agent

The Quick start assumes an interactive operator who answers the prompts and diagnoses a failure as it happens.

If you are an agent and the user has asked you to install Cartographer: fetch https://raw.githubusercontent.com/BeppeTemp/cartographer/main/docs/agent-install.md and follow it command by command. Do not install from the Quick start above: the runbook asks for the KB remote up front, verifies the install and ends with the session restart you cannot perform yourself.

For a human driving an agent, this is the prompt to paste:

Set up Cartographer on this machine by following
https://raw.githubusercontent.com/BeppeTemp/cartographer/main/docs/agent-install.md
My first knowledge base is at: `<git remote URL>`

One KB, every agent

A knowledge base is not only what an agent reads — it is also how that agent is set up to work. Cartographer treats skills, subagents, hooks and standing instructions as content of the KB, and materializes them into each client's native format.

The manual alternative is what most setups do today: the same skill hand-copied into .claude/skills/, .opencode/skills/ and .codex/skills/, each drifting on its own, each config file edited by hand for every MCP endpoint. Change one thing and you change it in six places, on every machine, forever.

cartographer connect        # detects installed clients and configures all of them

That single command writes, per client and in the format that client expects:

Subagents and hooks are translated, not copied: the same KB artifact becomes a Markdown agent for Claude Code, a TOML one for Codex, Antigravity-native Markdown, and a generated JavaScript plugin where a hook has no declarative equivalent.

Every — is an unsupported cell declared for a stated reason, never a silent omission — and a cell missing from the table fails a test:

• kiro — the shipped client has no hook mechanism that actually fires (verified empirically; details in docs/interoperability.md §Kiro hooks), so its re-sync trigger is the scheduled timer. Subagents work: ~/.kiro/agents/<name>.json is discovered globally and the built-in agent delegates to it by description.
• hermes — its MCP endpoints and its always-on instruction slot are rendered by its own Ansible role and recreated on the next playbook run, and it has no subagent directory and no hook engine. Skills are delivered, not installed: they land in an inbox with a generated SOURCE.md and the agent's own curator adopts them, because overwriting what that curator owns would destroy its learning.

Where there is no session hook, the scheduled trigger takes over (cartographer service sync-timer install): opt-in, explicit, and never installed as a side effect of connecting.

What keeps it true after the first run:

• it re-syncs by itself — a SessionStart hook on every client that has one, a scheduled timer for those that don't;
• it verifies the files on disk, not just its own bookkeeping: an artifact edited by hand or deleted is restored from the server;
• every materialized file carries a provenance stamp saying which KB it came from and where to edit it for real;
• it only ever touches what it created — pruning is limited to its own tracked paths, and --dry-run shows the plan without writing;
• doctor diagnoses residues and drift read-only; reconnect rebuilds a client from scratch while preserving every setting.

Edit a skill once in the KB, and every client of every machine converges on it.

Teams

The same mechanism is what makes Cartographer work for more than one person. A server mounts several KBs, each on its own endpoint (/mcp/<name> or ?kb=<name>), so colleagues can each keep a private knowledge base while sharing others. With mcp.mount_mode: routed the server also exposes a single /mcp/routed surface where the KB is a tool argument: a client using several KBs then carries one copy of the tool schemas instead of one per KB.

• Per-KB authorization — bearer tokens carry kb:<name>:r or kb:<name>:rw scopes. Roles (docs/transport-auth.md) narrow that further to specific maps, journals and concept types, so a teammate can be an editor of the runbooks and a reader of everything else. Rules are unioned: adding a role can only widen access, never silently revoke it.
• Git is the sync layer — every write is a commit, with fetch/pull-rebase before and push after, so teammates running their own server against separate clones of the same remote converge without a coordination protocol. A conflict is then not an error page but a workflow: the affected concepts are flagged degraded, conflicts_list enumerates them, and a bundled skill walks an agent through resolving them. (One process is the sole writer of a given working copy; pointing two writers at one checkout is not a supported model — partition KBs across instances instead, see docs/concurrency.md.)
• Shared content stays portable — a skill that mentions a local repository uses a {{repo:<name>}} placeholder resolved on each client from its own git remotes, so the same artifact works on every teammate's machine without machine-specific paths leaking into the KB. A server-side lint flags the ones that do.
• Provenance you can verify — a KB can sign its provisioning artifacts with Ed25519; clients pin the public key out of band and refuse anything that fails verification. Distributing a skill to a team is then a checkable act, not a matter of trust.
• Per-KB identity — the commit author is configured per KB, so history attributes correctly; on per-KB endpoints an optional tool-name prefix keeps an agent mounting several KBs from confusing their tools (the routed surface needs none: the KB is a call argument).

Key features

• 🔧 Full MCP tool suite — complete list in docs/control-plane.md
• 📖 Read & navigation — atlas_overview, index_get, concept_read, map_list, graph_neighbors (outbound links or backlinks), graph_context (ranked context around a question or concepts), link_suggest (links a concept probably lacks) and concept_list (scoped frontmatter facets)
• 🔍 Search — keyword: a pure-Go inverted index, or SQLite FTS5 with a trigram tokenizer when the KB has a persisted index; titles weigh most, and accents are ignored (citta finds città)
• ✍️ Validated writes with optimistic concurrency (if_match / content-hash), including concept_new from KB-owned templates discovered through template_list
• ✂️ Bounded edits and batches — concept_patch and index_patch apply Edit-like patches to a concept or a curated index.md; concept_batch makes a large refactor atomic across many concepts (one commit, full rollback on any failure)
• 📎 Concept assets — read, write, list, and delete binary or text dossier files inside expanded concepts
• 🛡️ Governance — deterministic lint (broken link, stale claim, orphan, map contracts), gate_check (validation + lint + commit gate in one call), supersede, contradiction tracking
• 🧬 Transactional git — one commit per write operation; optional synchronization to a remote (fetch/pull-rebase before and push after every write), which is also what lets several instances serve one KB — see Teams
• 🔐 Audit log — append-only with hash-chain and Ed25519 signature
• 🧩 Domain skills (SKILL.md / agentskills.io format), including executable scripts and binary assets — see One KB, every agent for how they reach each client
• 🔑 Secrets via SOPS — JSON Pointer references, scoped resolution and safe rotation; plaintext values never stored
• 🗺️ Embedded Atlas UI — a read-only web view of each KB served by the same binary at /ui/ in HTTP mode: a live 3D graph you orbit and select from, coloured by community or by Map, with an inspector and the lint findings; light and dark; no extra service, no CDN. Disable with web.enabled: false — see docs/deployment.md
• 📦 OKF-compliant — each KB is an OKF bundle and a standalone git repo, zero lock-in (just git + Markdown)

Architecture

Cartographer separates a data plane from a control plane:

• Data plane — the KB itself: OKF Markdown files under data/, organized as atlas → map → concept (the KB, its thematic archives, the pages; journals are the chronological maps). Plain files + git: history, diff, backup, sharing for free.
• Control plane — the MCP tools the agent calls. The server applies every invariant (validation, gates, immutability) so the agent operates safely without direct filesystem access.

The interaction rests on the MCP + Skill + Hook triad: MCP carries data and capabilities, Skills carry procedural know-how loaded on demand, Hooks carry deterministic 0-token automation.

flowchart LR
    A["🤖 Agent (LLM)<br/><i>only via MCP — never touches files</i>"]
    S["Cartographer<br/>Go MCP server<br/><i>invariants enforced server-side</i>"]
    KB[("KB<br/>Markdown + git")]
    R[("remote git")]
    A -- "MCP tools" --> S
    S -- "bounded reads" --> A
    S -- "one commit<br/>per write" --> KB
    KB -. "sync in/out" .-> R

Install

# macOS (Homebrew)
brew install beppetemp/tap/cartographer

# Windows (PowerShell; no administrator rights)
irm https://raw.githubusercontent.com/BeppeTemp/cartographer/main/install.ps1 | iex

# Linux / macOS without Homebrew (Darwin and Linux only — the script refuses on
# Windows and points at install.ps1)
curl -fsSL https://raw.githubusercontent.com/BeppeTemp/cartographer/main/install.sh | sh

# From source (Go 1.26+)
go install github.com/BeppeTemp/cartographer/cmd/cartographer@latest

What gets installed

• The binary, cartographer — in Homebrew's prefix (brew), in %LOCALAPPDATA%\Cartographer\bin, added to the user PATH (install.ps1: nothing is written outside your user profile), in /usr/local/bin or, when that is not writable, ~/.local/bin (install.sh), or in $GOBIN/$GOPATH/bin (go install).
• A native per-user service, if you run cartographer service install: ~/Library/LaunchAgents/com.cartographer.serve.plist on macOS, ~/.config/systemd/user/cartographer.service on Linux, or the Scheduled Task \Cartographer\Serve on Windows, listening on 127.0.0.1:39273. Its config is generated at ~/.config/cartographer/server.yaml — on Windows %APPDATA%\cartographer\server.yaml, with the log at %LOCALAPPDATA%\cartographer\Logs\server.log. None of the three needs administrator rights. The service is optional — a stdio-only setup (serve --kb <path>) is a legitimate topology and installs none of this.
• A data directory, ~/cartographer-data by default, holding the cloned KBs.
• Writes into your agent clients' own configuration under $HOME, and only when you run cartographer connect — never before. Each destination path is listed in the One KB, every agent matrix above. A sync timer (com.cartographer.sync / cartographer-sync.timer) is installed for clients that have no session-start hook.

Upgrades

Upgrades of a native local install repair themselves: install.sh update and install.ps1 update restart the running service on the new binary and re-synchronize the configured providers in place; brew upgrade runs no Cartographer code, so there the next cartographer sync — the session-start hook, the scheduled task, or a manual run — replaces a service still running the previous binary. On Windows the update lands even while the service is running from the file it replaces:

& ([scriptblock]::Create((irm https://raw.githubusercontent.com/BeppeTemp/cartographer/main/install.ps1))) update

You hear about a new release from your agent: at session start it is told once, with the upgrade command for your install channel, and offers to run it (cartographer update check asks directly).

cartographer reconnect is the explicit rebuild for what an incremental sync cannot see. Only already-open agent sessions need restarting. Details → docs/deployment.md §Upgrades, schema migration, and repo growth.

How to remove it

install.sh uninstall and install.ps1 uninstall remove the binary only. Both refuse to run while the native service or the sync timer is still installed, and name the teardown that has to come first:

cartographer disconnect                      # removes what was materialized into your agents
cartographer service sync-timer uninstall
cartographer service uninstall
curl -fsSL https://raw.githubusercontent.com/BeppeTemp/cartographer/main/install.sh | sh -s -- uninstall

cartographer disconnect
cartographer service sync-timer uninstall
cartographer service uninstall
& ([scriptblock]::Create((irm https://raw.githubusercontent.com/BeppeTemp/cartographer/main/install.ps1))) uninstall

brew uninstall checks nothing: run the same three commands before it.

Your KBs are git repositories in the data directory: nothing above deletes them, and removing ~/cartographer-data is a deliberate, separate act.

Configuration

The server reads a YAML file — cartographer service install generates ~/.config/cartographer/server.yaml, and config.example.yaml is the annotated example. Environment variables and CLI flags override it (flag > env > YAML > default). The ones most often set:

Full list with CLI flags and defaults → docs/deployment.md.

Building and testing

make build         # → bin/cartographer
make test          # Unit tests (go test ./...)
make smoke         # stdio smoke test
make smoke-http    # operator-level HTTP smoke test (creates temp KBs via curl)
make e2e           # deterministic HTTP/CLI end-to-end scenarios
make web-test      # Atlas UI frontend tests (needs Node; `make gate` does not)
make web           # rebuild the committed UI bundle after changing web/

The E2E suite drives the compiled binary through HTTP, CLI, filesystem and real temporary git remotes. It is deterministic, requires no model credentials and runs in CI. Full strategy → docs/testing.md.

Project structure

cmd/cartographer/   # single binary: server (serve), client (connect/status/sync/kb/service), TUI
internal/           # one package per concern: KB model, MCP server, search, git, auth, provisioning, …
docs/               # full documentation (docs/index.md is the map)
test/               # deterministic HTTP smoke and cross-component E2E tests

Package-by-package map, with what each one owns → AGENTS.md §Code map (kept next to the contributor instructions so there is a single copy to keep true).

Documentation

Browsable at beppetemp.github.io/cartographer — same content as docs/, rendered.

The full index lives in docs/index.md. Main entry points:

• docs/getting-started.md — from zero to a working wiki, step by step
• docs/overview.md — vision, guiding principles, architecture
• docs/data-plane.md — KB model, hierarchy, OKF
• docs/control-plane.md — Go server, MCP tool API
• docs/concurrency.md — single-writer, git sync, conflicts
• docs/deployment.md — topologies (local service / k8s / multi-server), backup, env vars

Contributing

Issues and PRs are welcome — see CONTRIBUTING.md for the build/test loop, the PR flow (squash-merge, conventional titles, docs updated in the same PR), and how to find your way around the codebase. Cartographer is a personal project maintained on a best-effort basis: no response-time SLA. For security reports, see SECURITY.md.

License

Released under the Apache License 2.0. See LICENSE.